A large synthetic dataset for machine learning applications in power transmission grids

Area of Science:

• Electrical Engineering
• Computational Science
• Data Science

Background:

• Power grids face increasing operational challenges due to the energy transition, operating closer to limits under volatile conditions.
• Real-time assessment of power grid safety, stability, and reliability is crucial but hindered by limited access to historical operational data.
• Machine learning approaches require substantial data, which is difficult to obtain for real-world power systems.

Purpose of the Study:

• To present a novel method for generating a large synthetic dataset of power injections for electric transmission grids.
• To develop an algorithm capable of creating arbitrarily large time series data for grid analysis.
• To provide a statistically validated synthetic dataset that mimics real-world power grid operational data.

Main Methods:

• Generation of synthetic power injection data using grid topology (line admittance) and generator information (location, type, capacity).
• Integration of aggregated power consumption data, such as national load data from ENTSO-E.
• Statistical validation of the generated synthetic datasets against actual real-world data.

Main Results:

• A comprehensive synthetic dataset of power injections for a continental European electric transmission grid model has been generated.
• The developed algorithm successfully creates large-scale, statistically representative time series data.
• The synthetic data exhibits statistical properties comparable to real-world power grid operational data.

Conclusions:

• The presented method offers a viable solution for overcoming data scarcity in power grid operational analysis.
• The synthetic dataset facilitates the development and testing of fast, real-time computational approaches for grid stability assessment.
• This work supports the advancement of grid reliability and safety evaluations in the context of the ongoing energy transition.